mlds_2019
Labs for MATH 728 D
Machine Learning and Data Science
Spring 2019

mlds_2019, the home page for a set of MATLAB laboratory exercises associated with the class MATH 728 D: "Machine Learning and Data Science", taught in Spring 2019 by Professor Wolfgang Dahmen, of the Mathematics Department at the University of South Carolina. The labs were written and presented by John Burkardt.

The class met for lectures on Tuesdays and Thursdays, from 1:15-2:30 pm in LeConte 121.

John Burkardt was available in LeConte 401, on Mondays, from 1:10-2:00pm, for students who wish to work on any of the lab exercises, or who have questions about the assigned projects.

The class included both regular lectures, and as a follow-up, classroom lab exercises in which students attempted to carry out tasks related to the lecture material, using MATLAB or Python. The following lab exercises were originally developed at the University of South Carolina, under the guidance of Professor Wolfgang Dahmen, in the spring of 2019.

This material was later extensively revised and reworked for an undergraduate class in machine learning, planned and developed at the University of Pittsburgh, in consultation with Professor Michael Schneier, in the fall of 2019.

Some of this material has since been further developed and modified for a class at the Missouri University of Science and Technology, in consultation with Professor Yanzhi Zhang, in the fall of 2020.

Class information and lecture notes:

• syllabus.pdf;
• Introduction.pdf;
• I-LA.pdf, Linear Algebra Basis and Functional Analysis Snippets;
• II-Probability.pdf, Probability Basics;
• III-Tail-Bounds.pdf, Tail Bounds;
• IV-High-Dim.pdf, High Dimensional Geometry;
• V-Classification.pdf, Classification;
• VI-LearningPrinciples.pdf, Learning Principles;
• svd_applications.pdf, Applications of the Singular Value Decomposition;

Homework exercises:
(Homework is intended as exercises for you to familiarize yourself with the course material. It will not be collected or graded. If you have questions about the exercises, these can be answered through email or at office hours.)

• homework01.pdf, linear algebra;
• solution01.pdf solutions to linear algebra homework;
• homework02.pdf, probabiity;
• solution02.pdf, solutions to probability homework;

Project #1: Regression with Linear Least Squares (Due March 19)

• project1.pdf;
• mexico.txt
• filip.txt
• rising.txt
• sinus.m
• sinus.py

Project #2: High Dimensional Sampling and Ranking (Due April 11)

• project2.pdf;
• large_sparse.txt
• medium_sparse.txt
• small_sparse.txt
• tiny_sparse.txt

Project #3: Perceptron, SVM, Multilinear Regression, Clustering (Due April 25)

• project3.pdf;
• generators.txt, 56 rows, 4 columns: index, rpm, vib, grade(0,1);
• jet_engines.txt, 56 rows, 4 columns: index, rpm, vib, grade(-1,+1);
• insurance_train.txt, 1000 rows, 7 columns: age, sex, bmi, children, smoker, region, charges;
• insurance_test.txt, 338 rows, 7 columns: age, sex, bmi, children, smoker, region, charges;
• faithful.txt, 272 rows, 2 columns: eruption length, pause length;

LAB #1: MATLAB

• lab01.pdf;
• scrambled.txt, a file to be loaded, describing a matrix whose rows and columns have been scrambled.
• correct.txt, a file describing the unscrambled matrix.
• geyser_data.csv, a file of observations of the Old Faithful geyser, including the lengths of eruptions, and quiet times. Several data values have been corrupted, and should be removed before plotting the rest.

LAB #2: Linear Algebra

• lab02.pdf;

LAB #3: Plotting

• lab03.pdf;
• bulgaria_data.txt, a file listing census years and population counts for Bulgaria.
• faithful_data.csv, a file of observations of the Old Faithful geyser, including the lengths of eruptions, and quiet times.
• price_data.txt, the month, the year, and average monthly prices for 11 consumer products, between February 2008 and February 2018.
• schoolyear_data.m, a MATLAB M file that lists countries and school year lengths in days.
• snowfall_data.txt, snowfall measurements at Michigan Tech, from 1890 to 2017.
• volcano_data.csv, data defining the local elevation near a volcano.

LAB #4: Probability

• lab04.pdf;
• normal_pdf.m, evaluates the normal probability density function.
• normal_samples.m, samples the normal probability density function.

LAB #5: Optimization

• lab05.pdf;
• ford_data.csv, a header line, then 23 records of (model year, mileage, selling price).

LAB #6: Linear Regression

• lab06.pdf;
• china_data.txt, 12 records of (year,GDP per person);
• basketball_data.csv, a header line, then 30 records of (index, height, weight, sponsorship, age);

LAB #7: Multilinear Regression

• lab07.pdf;
• insurance_data.csv, a header line, then 1338 records of (age, sex, BMI, children, smoker, region, charges). sex, smoker and region are "text" values;
• insurance_data.txt, a header line, then 1338 records of (age, sex, BMI, children, smoker, region, charges); sex, smoker, and region are numeric values.

LAB #8: Logistic Regression

• lab08.pdf;
• aircon_data.txt, 44 records of (humidity, temperature, comfortable?(1/0));
• admit_data.txt, 100 records of (english score, math score, admit?(1/0));

LAB #9: Clustering

• lab09.pdf;
• faithful_data.csv, a file of observations of the Old Faithful geyser, including the lengths of eruptions, and quiet times.
• pollution_data.csv, Various measurements related to air pollution in US cities. There is an initial header line. There are 41 records. There are 8 fields: "City name", "SO2 mg/cm", "Average Temperature F", "Manufacturing Plants", "1970 Population", "Average Wind Speed mph", "Average Precipitation inches", "Annual Precipitation days"
• scale_01.m, a function which shifts and rescales each column of an array to have minimum 0 and maximum 1;
• swim.jpg, an image using 256 colors;

LAB #10: Gaussian Mixture Models

• lab10.pdf;
• climate_data.xls, 1570 records, 2 columns of data. The data naturally falls into two clusters.

LAB #11: Principal Component Analysis

• lab11.pdf;
• glass_data.csv, Data about chemical composition of samples of glass. There is an initial header line. 214 records are stored; Each record includes 11 fields: "Index","Refractive Index","Na","Mg", "Al","Si","K","Ca","Ba","Fe","Class"; Class:1=building float glass,2=building nonfloat glass, 3=vehicle Float glass, 4=vehicle nonfloat glass, 5=containers, 6=tableware, 7=headlamps.
• casablanca.png a gray-scale image from the movie "Casablanca", 460 pixels wide and 360 pixels high.

LAB #12: Naive Bayes Classification

• lab12.pdf;
• card_data.csv Collection of 50 playing cards; There is an initial header line. There are 50 records; There are 4 fields: Index, Rank(1-13), Suit(1-4), Order(1-52);
• crash_data.csv computer crash reports. There is an initial header line. There are 20 records. There are 5 fields: index(1-20), OS(1=Linux,2=OSX,3=Windows), LANG (1=C,2=Python), Browser(1=Chrome,2=Explorer,3=Firefox,4=Safari), Crash (0=No,1=Yes).
• loan_data.csv Data for a fraud-detection system. There are 20 records. There are 5 fields: ID(1-20), Credit History ("none", "paid", "current", "arrears"), Guarantor ("none", "guarantor", "coapplicant" ), Accommodation ("own", "rent", "free"), Fraud ("true", "false");
• tax_data.csv Tax information. There is an initial header line. There are 15 records: There are 5 fields: Index(1-15), Refund("true", "false"), Status ("single", "married", "divorced", Income (in thousands), Cheating ("true", "false").

LAB #13: Markov Methods

• lab13.pdf;
• hyperlink_data.csv, Hyperlink directed adjacency matrix for 16 web pages. There are 16 records: There are 16 fields: 0 or 1 if page i links to page j.
• risk_data.csv, Regional adjacency matrix for the game of Risk. There are 42 records: There are 42 fields of 0 or 1: adjacency to region 1, region 2, ..., 42.
• risk_map.png, A map that displays the names and numbers of the 42 Risk regions.
• risk_names.csv, Region names for the game of Risk. There is one initial record. There are 42 records: There is 2 fields: index, region name

LAB #14: Facial Recognition

• lab14.pdf;
• angela.zip 12 images of Angela Merkel, that are an example of a starting collection for the exercise;

LAB #15: Vector and Matrix Norms

• lab15.pdf;

LAB #16: Curve Fitting

• lab16.pdf;
• sine_test.txt 100 samples, at uniformly spaced points in [0,1], of the function y=sin(2*pi*x)+0.1*randn(), used for testing.
• sine_train.txt 10 samples, at uniformly spaced points in [0,1], of the function y=sin(2*pi*x)+0.1*randn(), used for training.
• spring.txt records the results of Hooke's law experiments, in which a spring is loaded with a given weight M, and the resulting deflection D is measured. A linear relationship D=c*M is expected. There are 20 records, with two fields, mass in kilograms, and deflection in meters. From Guttag, 2016.

LAB #17: Projection

• lab17.pdf;

LAB #18: Expected Values

• lab18.pdf;